Location Management Schemes

Location Management: Context

Mobility Management: Enables users to support mobile users, allowing them to move, while simultaneously offering them incoming calls, data packets, and other services.

Types of mobility:1. Terminal mobility: ability of terminal to retain connectivity with the network so that

all on-going communication services remain active despite terminal’s migration. 2. Personal mobility: disassociates user from the terminal (e.g. in GSM a mobile

station = mobile terminal + smart card with subscriber identification module (SIM)).3. Service mobility: provides continuous service to mobile clients across multiple

administrative domains. Consists of:

1. Location management: tracking mobiles and locating them prior to establishing incoming calls (deliverying pending messages).

2. Handoff management (a.k.a. automatic link transfer): rerouting connections with minimal degradation of QoS.

Location Management Problem

In static networks, a terminal’s network address serves two purposes:1. End-point identifier2. Location identifier

Mobility prevents using a single address for both purposes Both end-point identifier and location identifier are needed. Location management keeps mapping between an end-point identifier and its

location identifier Basically a directory problem.

Two primitive operations:1. Lookup (a.k.a. search/find/paging/locating) operation: is the procedure by

which the network finds the location of the mobile. required when a call (message) to a user is placed (to be delivered)

2. Update (a.k.a tracking/move/registration) operation: is the procedure by which the network elements update information about the location of the mobile. required when a user changes its “location” The information gathered during updating/tracking is used during the locating

operation

Location Management: Issues

More precise location needs to be maintained as cell size shrink: Wide area cells are 10’s – 100’s km in diameter Macro-cells: 1-10 km Micro-cells: 100’s m Pico-cells: under 10 m

Database issues in tracking mobile users: Maintaining update intensive location information Strategies to reduce location query latency (such as replication)

and traffic (such as caching) Consistency between replicas; Cache management polices

Location Management: Schemes

Several schemes have been developed which are motivated by fundamental trade-off between search operation cost and update operation cost.

Schemes which try to minimize one cost tend to increase the other cost Try to optimize the aggregate cost or normalized cost.

Categorization:1. Update Scheme: Static or Dynamic

Static update scheme: registration areas Dynamic update scheme: distance/time/movement based strategy

2. Locating Scheme: Static or Dynamic Static location scheme: page all the cells in the network Dynamic location scheme: expanding ring search centered at last reported location of

the the user3. Database Architecture: Flat or Hierarchical

Selection of LM Schemes

Cost of location updates and lookups Maximum service capacity of each location

database = the maximum rate of updates and lookups that each

database can service Space restrictions (size of the location database) Type and relative frequency of call to move

operations (call-to-mobility ratio (CMR))

One-Tier Scheme

A home database, called Home Location Register (HLR) is associated with each mobile user.

The HLR of a user x maintains the current location of x as part of x’s profile.

To locate a user x, x’s HLR is identified and queried.

When a user x moves to a new cell, x’s HLR is updated.

Two-Tier (Basic) Scheme

Visitor Location Registers (VLRs) are maintained in each zone (registration area).

VLR in a zone stores copies of profiles of users not at home and currently located in that zone.

When a call is placed from cell i to user x, the VLR at cell i is queried first, and only if the user is not found there, is x’s HLR contacted.

When user x moves from cell i to j, in addition to updating x’s HLR, the entry of x is deleted from VLR at cell i, and a new entry for x is added to the VLR at cell j.

Two-Tier Scheme: Standards

Many current and proposed standards use this scheme:Electronics Industry Association

Telecommunications Industry Associations (EIA/TIA) Interim Standard 41 (IS 41) - commonly used in North America.

Global Systems for Mobile Communications (GSM) - used in Europe.

Internet Engineering Task Force (IETF) Mobile IP protocol

Enhancements to Basic Scheme Per User Location Caching (dynamic replication) [Jain et.

al., IEEE JSAC 12(8), ‘94] reduces search (lookup) cost increases update cost exploits locality in call pattern

(Static) Replication Per User Profile Replication [Shivakumar & Widom Mobicom’95] Working Set Replication [Rajagopalan et. al., Mobicom’95]

Forwarding Pointers [Jain & Lin, Wireless Networks 1, ’95] reduces update (move) cost

Per User Location Caching

Basic Idea: Every time user x is called, x’s location (or a pointer to this

location) is cached at the VLR in the caller’s zone. Any subsequent call to x originating from that zone can use this

information: Upon call origination the cache at the VLR of the caller’s zone is

checked before querying the callee’s HLR.

Issues: cache replacement schemes (LRU can be used) cache invalidation schemes

eager caching or lazy caching

Eager and Lazy Caching

Eager Caching Every time a user moves to a new location, all cache entries for

this user’s location are updated. The cost of move operations increases for those users whose

address are cached. Lazy Caching

the cached pointer for any given user is updated only on a cache miss

for lazy scheme to work better than basic scheme p CH/CB where p is the hit ratio, CH is the cost of a lookup when there is a hit and CB is the cost of lookup in the basic scheme.

Replication

To reduce the lookup cost, the location of specific users is replicated at selected sites.

Let : cost savings when local lookup succeeds as

opposed to a remote query, : cost of updating a replica, Ci,j: expected number of calls made from cell j to i in a

unit time. Ui: expected number of moves by i in unit timeThen a replication of the location of user i at cell j is judicious if

Ci,j Ui.

Per User Profile Replication

Objective: to minimize the total cost of moves and calls, while maintaining Constraint 1: a maximum of ri replicas for user i, and

Constraint 2: a maximum of pj replicas in the database of cell j.

Replication assignment problem: The profile of user i is replicated at all cells in set R(i) such that the system cost N

i=1Mj=1,jR(i) (Ui- Ci,j)

is minimized, where N is the number of users and M is the number of cells., and constraints 1 and 2 above are met.

Replication Assignment

Flow Network based solution:

Solving Min-Cost Max-Flow on the Flow Network finds the required assignment.

s t

users cells1

i

N

(0,r1)

(0,ri)

(0,rN)

1

M

j (0,pj

)

(0,p1)

(0,pM)

(Ui- Ci,j,1)

(UN- CN,M,1)

Edges arelabeled with(cost,capacity)tuple.source sink

Working Set (WS) Replication

Relies on the observation that each user communicates frequently with a small number of sources called its working set.

Copies of location are maintained at the members of its working set.

No constraints are placed on database storage capacity or on number of replicas per user.

Hence, the decision to provide the information of the location of a mobile unit i to zone j can be made independently for each user.

Adapts to user’s call and mobility patterns

Working Set Adaptation

The inequality Q: Ci,j Ui. is evaluated locally at a mobile unit I each time:1. a call is set up,2. the mobile unit moves.

In case 1, Q is evaluated only if the caller’s site is not a member of mobile’s WS: If the inequality holds then the caller’s site becomes member of

the callee’s working set. In case 2, the Q is evaluated for every member of WS;

the members for which Q no longer holds are dropped from WS.

Performance of WS Replication

Computation overhead: in case 1 all four terms of Q need to be reevaluated in case 2 only the number of moves (Ui) needs to be

reevaluated. Adaptability:

when call-to-mobility ratio (CMR) value is low the WS scheme performs like a scheme without replication.

when CMR value is high, the scheme behaves like a static scheme in which the WS for a user is fixed.

Performance is mainly dependent upon the CMR of individual users (not on num. of users).

Forwarding Pointers

Each time a mobile unit x moves to a new location, a forwarding pointer is set up to its previous VLR to point to the new VLR.

To establish a call, the HLR of callee is queried to find the first VLR in the forwarding pointer chain. This chain is followed to get to the current VLR of the callee.

To bound the time taken for lookup procedure, the length of the chain is bound to a max value of K.

Pointer compression is used to eliminate loops.

Forwarding Pointers (Cont.)

Mobile IP protocol includes pointer forwarding in conjunction with lazy caching.

The forwarding pointer strategy is useful for those users who receive calls infrequently relative to the rate at which they change registration areas.

Benefits of forwarding depends also upon the cost of setting up and traversing pointers relative to the costs of updating the HLR.

Overlapping Registration Areas

Without overlapping

With overlapping A

B C

CB

A

12

3

12

3

Inter-RA hand-off: a user changes cells and RAs

Intra-RA hand-off: a user changes cells within an RA.

Inter-RA hand-off doesn’t happen as long as the hand-off can be intra-RA.

Inter-RA call is when caller and callee are in separate RAs

Intra-RA call is when caller and callee are in same RA.

A non-overlapping cell is serviced by one LR.

A overlapping cell is serviced by multiple LRs. Reduction of inter-RA hand-offs.

Overlapping RAs (cont.)

Advantages: Each RA can provide service to more mobiles within their

covered area. Reduces the number of inter-RA handoffs Reduce the load to update mobile’s HLR.

Disadvantages: the communication overhead for call-delivery and intra-RA

handoff is increased. the increase in overhead depends upon the underlying

network topology. If this overhead is ignored then the extreme configuration in

which each RA has all the cells in the system becomes the “optimal” configuration.

Overlapping RAs (cont.)

Dynamically Resizing RAs We need to find optimal configuration (allowing

overlapping RAs) i.e. configuration which minimizes load on MSSs.

When move and call patterns periodically change, a static scheme may not provide a good solution

Our Approach: Allow RAs to be dynamically adapted. Periodically resize RAs to minimize MSS load:

Resizing criterion: load reduction due to lesser number of inter-RA handoffs > increase in load due to more expensive call delivery and intra-RA handoffs.

If resizing criterion is ignored then each RA will grow to maximum size.

Overlapping RAs (cont.)

Even though mobiles a and b belong to the same RA, anycalls between them would need to go through two MSSs.

Backbone Network

MSS/LR

Cells

RA BRA A

RA Ca b

Negative effect of underlying conventional star topology on signaling overhead under overlapping RAs

Overlapping RAs (cont.)

Inclusion and Exclusion Boundary In order to facilitate

orderly growth and shrinking of RAs, an MSS only includes and excludes cells from its RAs current boundary.

Two types of boundary:1. Internal Boundary2. External Boundary

MH BS

MSS

Overlapping RAs (cont.) Inclusion/Exclusion Decision

The decision to include or exclude a candidate cell is based on whether the resulting configuration will have a lower expected load on MSS.

For a given system configuration A, mobility pattern M, and call C, SystemLoad(A,M,C) is the combined signaling load (in terms of message time complexity) as a result of all the handoffs due to M and call-deliveries due to C: SystemLoad(A,M,C) = Load(k,M,C).

In case of inter_RA handoffs and call-deliveries we spilt the signaling overhead equally between the two MSSs involved.

Overlapping RAs (cont.) What changes when cell x is included in RA r?

Handoffs to cell x from cells of RA r become intra-RA handoffs. Handoffs from cell x to rest of RA r performed by users already

registered in r become intra-RA handoffs. Calls to x from cells of r are now intra-RA calls. Calls from users of r that are in x to rest of r are now intra-RA

calls. Mobility of users in r that move out of cell x into a new RA is now

inter-RA mobility. Inter-RA calls of users in r that call from cell x is inter-RA call

loading to r. Call the decreasing part of the load Costin(x,r) and the

increasing part Costex(x,r). At intervals T each MSS/LR r computes I_Boundary(r) and

E_Boundary(r) and for each cell x in the two sets computes Costin(x,r) and Costex(x,r). By comparing the two values, it decides if it is worth keeping excluded, keeping included, including or excluding the cell c.

Dynamic Updating

Most schemes have fixed locations (i.e. the boundaries of registration areas) where the mobiles update.

Users that move around boundaries cause a lot of registrations.

[Bar-Noy95] Solution: introduce dynamic update schemes that don’t depend on location of mobile.

Dynamic Updating (cont.) Time-based User updates location at intervals of time T

independent of actual location. Movement-based

User updates location after crossing M hops (cells) from last updated location.

Distance-based User updates location after being distance D from

last updated location. Two metrics to evaluate schemes:

update rate (# of updates/sec) search area (# of cells/search)

Dynamic Updating (cont.) Time-based versus Movement-

Based Update rate

If user crosses less than M cells per time T, then time-based makes more updates, otherwise time-based makes less updates

If M is average hops traversed per time T, then two schemes have same rate of updates.

Search area Search area in time-based is the

cells that can be reached from last updated location at max user speed in time T.

Search area in movement-based is cells that can be reached in M or less hops from last updated location

Dynamic Updating (cont.)

Movement-based versus Distance-based

Update rate If D=M, distance-based will

do at most as many updates as movement-based

Search area Search area is same in both

schemes (cells that can are at distance D or less from last updated location)

Look-ahead update [Tsai97] Proposed a look-ahead update scheme based on the distance-based scheme of [Bar-Noy95].

Mobility is modeled as a “normal walk” where the mobile tends to keep the direction of movement

Look-ahead update scheme: in distance-based scheme don’t update current location, but update a location Δ hops ahead

Under normal walk mobility model, user is more probable to cross standard circle (solid) before crossing look-ahead circle (dotted). Therefore look-ahead saves updates.

Dynamic Hierarchy [Ho97] Proposed a hybrid scheme with

different hierarchy levels. Directory Registers: are a inter-mediate

level of hierarchy between VLRs and HLRs.

For each user there is the consept of Local DR, the DR that is above the VLR.

HLRs may point to VLRs (direct pointer) or to LDRs (indirect pointer). Hence, the two hierarchies:

Two levels: HLRVLR Three levels: HLR LDR VLR

Scheme includes Forwarding Pointers Of users various DRs to other DRs or LRs

Local DR

MSCa

Remote DR1

Remote DR2

MSCb

MTa MTb

HLRa HLRb

Indirect HLR pointer

Direct HLR pointer

Direct DR remote pointer

Indirect DR remote pointer

Multiple-level Hierarchical Scheme [Bejerano98] define a multiple-

level hierarchy of overlapping Location Areas: If LAs are viewed as circles of

radius r, then the outer r/2 part (periphery) overlaps with the inner parts (cores) of neighboring LAs, and the inner r/2 part (core) overlaps with the outer parts (peripheries) of neighboring LAs.

At each level n, there is twice the number of the (n+1)-level Location Areas and half of the (n-1)-level Location Areas.

Logarithmic number of levels For every user, there is at least

one LA at each level that has a “location pointer” to a LA to the next lower level.

Multiple-level Hierarchical Scheme (cont.)

Update policy: At each level, starting from lowest, if the

user moves between two cells that are not in the same LA, the move is “updated” to the LA in the above level as well.

A movement update goes up to the LA that embraces both ending and starting cells of the user’s movement.

Search Policy: If there is no downward pointer, then the

search is propagated upward until a LA has a downward pointer of the user.

The downward pointers are followed until the user is reached.

Scheme shows very good average and best case costs, but very bad worst case.

Summary of Variations to 2-Tier Scheme

Method Variations Used When:

CachingWhen x is called by y, cache

x’s location at y’ zone.

Eager caching

Cache update overhead occurs at moves Large

LCMRLazy Caching

Cache update overhead occurs at calls

ReplicationSelectively replicate x’s address at the zones from which it receives the most calls

Per-user profile Replication

Additional constraints are set on the number of replicas per site and on the number of replicas per user Large

LCMRWorking Set

Adaptive Distributed: the replication sites are computed locally at each mobile host

Forwarding PointersWhen x moves, add a forwarding pointer from its old to its new address.

Restrict the length of the chain of forwarding pointers

Small LCMR

Hierarchical Schemes (HS)

Extend two-tier schemes by maintaining a hierarchy of location databases.

Location database at higher level contains location of users located at levels below it.

Usually hierarchy is tree structured: Location database at a leaf serves a single cell and contains

entries for all users registered in that cell. A database at an internal node maintains location of users

registered in the set of cells in its subtree. location information can be either

pointer to an entry at a lower level database or the user’s actual current location.

Hierarchical Location Scheme

0

1 2

5 63 4

7 8 9 10 11 12 13 14 15 16 17 18

user x

x

x

x 9

entries for user x whenthe location databasemaintain pointers

x 9

x 9entries for user x whenthe location databasemaintain actual location

x 9

x x 9

[Jain, ICC ’96]

Updates/Lookups with Pointers

LCA(i,j): least common ancestor of i and j. When user x moves from cell i to j, following entries for x

in databases are updated:1. along the path from j to LCA(i,j), and 2. along the path from LCA(i,j) to i.

When a caller located at cell i places a call for a user y located at cell j, the lookup procedure: 1. queries databases starting from node i and proceeding upwards

the tree until the first entry for x is encountered (at LCA(i,j)). 2. Then the lookup procedure proceeds downwards following the

pointers to node j.

Update/Lookup Example

0

1 2

5 63 4

7 8 9 10 11 12 13 14 15 16 17 18

user x(moved from cell 9to cell 10)

x

x

x 9

x

LCA(9,10)

x

x 10

upward phase:till entry for calleeis found

downward phase:follow the pointer

LCA(10,13)

a node in cell 13calls user x

Lookup/Update with Actual Location

When user x moves from cell i to cell j: record for x is deleted from all the databases from node i to

LCA(i,j), and record for x is updated to indicate the current location to be cell j

in all the databases from root node to leaf node j. When a user x from cell i places a call to user y in cell j

the lookup procedure queries database at node i proceeding upwards till node LCA(i,j).

Compared to pointers case, in this case:1. updates are more expensive operation, and2. lookups are less expensive operation.

Update/Lookup Example

0

1 2

5 63 4

7 8 9 10 11 12 13 14 15 16 17 18

user x(moved form cell 9 tocell 10)

x

x

x 9

x

LCA(9,10)

x

x 10

9

910

910

10

a node in cell 13calls user x

LCA(10,13)

Caching in Hierarchical Scheme0

1 2

5 63 4

7 8 9 10 11 12 13 14 15 16 17 18

s

t

forward bypass pointer

reverse bypass pointer

user y (callee)

search procedure

user x (caller)•Forward bypass pointer is an entry at an ancestor of caller’s cell, say s, that points to an ancestor of callee’s cell, say t.•The reverse bypass pointer is from t to s.•In simple caching both s and t are leaf nodes. •In level caching s and t can belong to any (possibly different) levels.

Forwarding Pointers in Hierarchical Scheme

Reduces the update cost in case of move from cell i to cell j, instead of updating all databases on the path from j through LCA(i,j) to i, only the databases up to a level m are updated and a forwarding pointer is set from a node s to node t, where s is the ancestor of i at level m and t is ancestor of j at level m.

0

1 2

5 63 4

7 8 9 10 11 12 13 14 15 16 17 18

user x new location

new enteries

old enteries

user x old location

level m Forwarding Ptr

Lookup cost for calls initiated from any cell inthis subtree is increased

Hierarchical Schemes: Summary

Method Description

Caching

When x at zone i is called by user y at zone j, cache at a node on the path from j to LCA(i,j) a pointer to a node on the path from I to LCA(i,j) to be used by subsequent calls to x from zone j.

Replication

Selectively replicate x’s location at internal and/or leaf database.

Forwarding Pointers

When x moves from cell i to cell j, instead of updating all databases on the path from j to LCA(i,j) and from LCA(i,j) to j, update all databases up to level m and add a forwarding pointer at the level m ancestor of I to point to the level m ancestor of j.

Hierarchical vs. Two-Tier Scheme

1. No pre-assigned HLR

2. Support Locality

3. Increased number of operations (database operations and communication messages)

4. Increased load and storage requirements at the higher-levels

Location Management: Summary Location management is a rich research topic;

the following represents 3D space of possible solution [ What(granularity), Where (availability), When (currency)]:

Ava

ilab

ilit

y

nowhere

at all sites

At selective sites (e.g., at frequent callers)

CurrencyNever update

Always update (at each movement)

Granularity

Exact location

the whole network

some partition

References

[Akyildiz97] J. S. M. Ho and I. F. Akyildiz. Dynamic Hierarchical Location Management in PCS Networks. IEEE/ACM Transactions on Networking, 5(5):646–660, October 1997.

[Bejerano98] Y. Bejerano and I. Cidon. An Efficient Mobility Management Strategy for Personal Communication Systems. pages 215–222, MobiCom 98, April 1998.

[Bar-Noy95] A. Bar-Noy, I. Kessler, M. Sidi, Mobile Users: To Update or not to Update? ACM/Baltzer Wireless Networks Journal. Vol 1, no 2, 1995, pp. 175-186.

[Tsai97] I-F Tsai, R-H Jan, The Lookahead Strategy for Distance-Based Location Tracking in Wireless Cellular Networks, Mobile Computing and Communications Review, Vol 3, No 4, 1997